A method and system for generating intelligent test questions based on learning behavior analysis

Abstract

This invention relates to the technical field of test question generation, and in particular to an intelligent test question generation method and system based on learning behavior analysis. The method includes: acquiring real-time interaction behavior data and performance data of learners in an online learning platform; inputting the interaction behavior data and performance data into a pre-trained knowledge state analysis model to generate a user knowledge state matrix; based on the knowledge state matrix and a preset teaching objective library, identifying the set of target knowledge points that need to be strengthened and the corresponding cognitive training types; and, according to the set of target knowledge points that need to be strengthened and the corresponding cognitive training types, calling a test question element combination algorithm to dynamically assemble the question stem, distractors, and solution path hints to generate personalized test questions. This application has the effect of realizing a fully closed-loop intelligent teaching system from user behavior analysis to targeted training, and dynamically generating personalized test questions adapted to individual cognitive gaps.

Description

Technical Field

[0001] This invention relates to the technical field of test item generation, and in particular to an intelligent test item generation method and system based on learning behavior analysis. Background Technology

[0002] With the development of intelligent education technology, personalized test question generation has become a core means to improve learning efficiency. Traditional adaptive learning systems mainly rely on shallow behavioral data such as student answer accuracy and answer time, generating entities through item response theory or knowledge graph matching. However, this type of test question generation method has the following drawbacks:

[0003] 1. One-sidedness of cognitive assessment: Relying solely on the answer results data while ignoring deeper cognitive behaviors such as attention allocation and strategy adjustment during the problem-solving process leads to a high bias rate in knowledge state modeling;

[0004] 2. Lack of knowledge point association: Existing technology treats knowledge points as independent units and fails to capture the topological dependencies between knowledge points. When users encounter association errors, the root cause cannot be located.

[0005] 3. Mechanized generation of distractors: The rules for generating distractors in the question bank are often manually preset, lacking dynamic adaptation to user error patterns.

[0006] Therefore, there is room for improvement. Summary of the Invention

[0007] To achieve fully closed-loop intelligent teaching from user behavior analysis to targeted training, and to dynamically generate personalized test questions that adapt to individual cognitive gaps, this application provides an intelligent test question generation method and system based on learning behavior analysis.

[0008] The above-mentioned objective of this application is achieved through the following technical solution:

[0009] A method for generating intelligent test questions based on learning behavior analysis, the method comprising the following steps:

[0010] Real-time acquisition of learner interaction behavior data and performance data in the online learning platform, wherein the interaction behavior data includes the duration of time spent on knowledge points and the behavior of jumping to and from learning videos, and the performance data includes answer records and the frequency of incorrect questions;

[0011] The interactive behavior data and performance data are input into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, which includes the degree of mastery of knowledge points, the type of wrong questions, and the cognitive load index.

[0012] Based on the knowledge state matrix and combined with the preset teaching objective library, identify the set of target knowledge points that need to be strengthened and the corresponding cognitive training type.

[0013] Based on the set of target knowledge points to be strengthened and the corresponding cognitive training types, the test element combination algorithm is invoked to dynamically assemble the question stem, distractors, and solution path hints to generate personalized test questions.

[0014] By adopting the above technical solution, multi-dimensional interactive behavior data (including knowledge point dwell time and video viewing / jumping behavior) and structured performance data (such as answer records and frequency of incorrect questions) from users' online learning platforms are captured in real time, constructing a fine-grained learning behavior profile. This solves the problem of traditional assessment relying on a single answer result. Furthermore, the fused data is input into a pre-trained knowledge state analysis model, which uses temporal behavior pattern recognition and error clustering algorithms to generate a user knowledge state matrix containing quantifiable values ​​of knowledge point mastery (accurate to the 0-1 range), error type classification labels (such as conceptual confusion / calculation error / logical gap), and a cognitive load index (calculated based on attention fluctuation characteristics). This significantly improves the accuracy of cognitive state assessment and overcomes the limitations of traditional assessment methods. Existing technologies neglect the implicit learning process. Based on the matrix, the achievement threshold of the teaching objective library is dynamically matched. The set of target knowledge points that need to be strengthened is located through knowledge gap topology analysis. The cognitive training type is adaptively selected according to the type of wrong questions (such as adversarial training for conceptual confusion and transfer training for logical gaps), realizing an intelligent decision-making closed loop from "knowledge gap detection" to "training strategy generation". Finally, the test question element combination algorithm is called to dynamically assemble the semantic framework of the question stem according to the core examination elements of the target knowledge point set. The deceptiveness of the interference items is optimized by using Generative Adversarial Network (GAN), and the problem-solving path prompts that are accurately matched with the user's historical error patterns are injected (such as inserting a knowledge point definition comparison table for conceptual confusion errors), generating completely personalized test questions.

[0015] In a preferred embodiment, this application can be further configured such that: the real-time acquisition of user interaction behavior data and grade data in the online learning platform specifically includes:

[0016] Based on the data collection logs, the number of times users modify their answer choices and the final submission delay are captured. The user-marked confusion tags and related notes in the error notebook are recorded and integrated to form the performance data.

[0017] Extract pause / speed playback segments from video learning and their associated knowledge point tags to generate interactive behavior data;

[0018] The performance data and interaction behavior data are aggregated to form a time-stamped sequence of user behavior vectors.

[0019] By adopting the above technical solutions, the dynamic decision-making behavior of users during the question-answering process (including the number of option modifications and the final submission delay) is accurately captured through the tracking of logs. The confusion marks and related notes self-marked by users in the error notebook are recorded simultaneously, extending traditional performance data to the cognitive decision-making level. This upgrades error attribution analysis from simple result judgment to process tracking (e.g., modification count > 3 times and delay > 2 standard deviations are marked as "conceptual hesitation type" errors), significantly improving the accuracy of error pattern recognition. At the same time, pause / speed playback segments in video learning are extracted and bound to knowledge point tags (e.g., pause rate > 80% in key formula derivation segments is automatically associated with weak knowledge points). Unstructured interaction behavior is transformed into quantifiable attention allocation indicators. By making full use of video learning data, the above performance data and interaction behavior data are aggregated to generate user behavior vector sequences with millisecond-level timestamps. Multidimensional heterogeneous data are integrated through time-series alignment technology to construct a three-dimensional behavioral profile covering "question-answering decision chain - video focus - self-reflection text", providing a high-fidelity data foundation for subsequent knowledge state modeling.

[0020] In a preferred embodiment, this application can be further configured such that: inputting the interaction behavior data and performance data into a pre-trained knowledge state analysis model to generate a user knowledge state matrix specifically includes:

[0021] The timestamped user behavior vector sequence is decoupled to separate knowledge proficiency features, attention features, and metacognitive strategy features.

[0022] The knowledge proficiency features, attention features, and metacognitive strategy features are input into a pre-trained knowledge state analysis model to construct a topological correlation strength matrix between knowledge points.

[0023] Based on the topological correlation strength matrix between knowledge points and the user's historical answer accuracy, the quantitative indicators of each dimension in the topological correlation strength matrix between knowledge points are calculated to generate the knowledge point mastery and cognitive load index.

[0024] Error pattern classification is obtained based on the user's historical answer accuracy rate. The type of wrong question is determined based on the error pattern classification. The user's knowledge status matrix is ​​generated based on the knowledge point mastery, cognitive load index and wrong question type.

[0025] By employing the aforementioned technical solution, multi-dimensional feature decoupling is performed on the timestamped user behavior vector sequence to separate knowledge proficiency features (calculated based on the exponential decay weighted calculation of answer accuracy), attention features (quantified by the coefficient of variation of video pause intervals), and metacognitive strategy features (using BERT clustering analysis of incorrect question annotation text). These decoupled features are then input into a pre-trained knowledge state analysis model, which uses a graph neural network to dynamically construct a topological association strength matrix between knowledge points (e.g., the association strength between trigonometric functions and vector knowledge points). This accurately captures implicit knowledge dependency chains, addressing the deficiency of static knowledge graphs in reflecting individual cognitive differences. Based on this topological matrix, historical answer accuracy is fused to generate dual-channel quantitative indicators: knowledge point mastery and cognitive load index. Simultaneously, an error pattern classifier is constructed based on historical answer data. Through confusion matrix analysis (e.g., similar option error rates > 40% are marked as conceptual confusion), incorrect question types are refined into actionable training targets. Finally, these are integrated into a user knowledge state matrix, providing an irreplaceable cognitive map for subsequent target knowledge point positioning.

[0026] In a preferred embodiment, this application can be further configured as follows: based on the knowledge state matrix and combined with a preset teaching objective library, identifying the set of target knowledge points that need to be strengthened and the corresponding cognitive training type specifically includes:

[0027] By comparing the knowledge status matrix with the achievement thresholds in the teaching objective database, the knowledge points that have not been achieved can be located.

[0028] Based on the types of incorrect questions in the knowledge state matrix, select the training type for each knowledge point, and determine the training intensity level based on the cognitive load index in the knowledge state matrix.

[0029] Based on the knowledge point training type and training intensity level, the output is a set of required instructions that includes the target knowledge point ID, training type, and intensity level.

[0030] By adopting the above technical solution, the mastery of knowledge points in the knowledge status matrix is ​​intelligently compared with the preset threshold of the teaching objective library (such as the mastery of the function chapter in junior high school mathematics ≥0.8). The unmet knowledge points are accurately located, and then the cognitive training type is dynamically selected based on the detailed error types in the matrix. This achieves a precise match between the root cause of the error and the correction strategy. At the same time, the training intensity level is determined by combining the cognitive load index, and the complexity of the test questions is dynamically adjusted (such as the number of implicit conditions in the question stem increases or decreases linearly with the intensity coefficient). This effectively avoids the risk of cognitive overload and finally outputs a structured demand instruction set.

[0031] In a preferred embodiment, this application can be further configured as follows: based on the set of target knowledge points to be strengthened and the corresponding cognitive training type, the step of calling a question element combination algorithm to dynamically assemble the question stem, distractors, and solution path hints to generate personalized questions specifically includes:

[0032] Based on the target knowledge point ID of the aforementioned requirement instruction set, the core examination elements are matched from the test question material library;

[0033] Based on the training type, a distractor generation strategy is selected, and in response to the distractor generation strategy, question distractors are output.

[0034] The complexity parameter of the test questions is determined based on the training intensity level. Based on the complexity parameter and the distractors, the solution path hints that match the user's wrong question type are injected to generate the final personalized test questions.

[0035] By adopting the above technical solution, based on the target knowledge point ID of the demand instruction set, the core examination elements are intelligently matched from the test question material library (such as binding the three elements of vertex form, axis of symmetry, and discriminant form to the knowledge point of "extreme value of quadratic function") to ensure the targeting of test questions. The strategy of generating distractors is dynamically activated for different training types. For example, adversarial training calls the confusion concept library to generate distractors with semantic similarity > 85% (such as replacing "extreme point" with "stationary point"). Transfer training maps to cross-disciplinary scenario templates (such as using physical kinematics cases to package mathematical function problems) to increase the deceptiveness of distractors. The complexity parameters of test questions are precisely adjusted according to the training intensity level to effectively balance the training challenge. Finally, the solution path prompts that strictly match the user's wrong question type are injected to generate fully structured personalized test questions.

[0036] In a preferred embodiment, this application can be further configured such that: the interference item generation strategy based on training type outputs question interference items in response to the interference item generation strategy, specifically including:

[0037] The training types include adversarial training and transfer training. In adversarial training, distractors with similar semantics but different knowledge point affiliations are generated based on confusing concepts.

[0038] In transfer training, core knowledge points are mapped to preset interdisciplinary scenario templates to generate distractors;

[0039] By optimizing the persuasiveness weights of distractors and correct answers using an adversarial generative network, low-persuasiveness distractors are filtered out based on their weights.

[0040] By adopting the above technical solutions, in adversarial training, distractors with semantic similarity >85% but different knowledge point affiliations are generated based on a confusing concept library (such as "stationary point" and "extreme point" in mathematics, and "electrolyte" and "non-electrolyte" in chemistry). The cosine similarity of word vectors is used to ensure the cognitive confusion between the distractors and the correct answer, and to specifically solve conceptual confusion errors. In transfer training, contextualized distractors are generated by pre-set cross-disciplinary scenario templates (such as mapping mathematical derivatives to instantaneous velocity cases in physics, and transferring chemical equilibrium to ecological population competition models), which improves the recognition rate of abstract knowledge points. Then, adversarial generative networks are used to dynamically optimize the deceptiveness weight of distractors, calculate the deceptiveness score based on historical answer data, and filter out distractors with low deceptiveness.

[0041] In a preferred embodiment, this application can be further configured as follows: after the step of dynamically assembling the question stem, distractors, and solution path hints by invoking a question element combination algorithm according to the set of target knowledge points to be strengthened and the corresponding cognitive training type, and generating personalized questions, the intelligent question generation method based on learning behavior analysis further includes:

[0042] Obtain user response datasets and problem-solving process data for the generated personalized test questions;

[0043] By comparing the deviation between the actual answer path and the expected answer path, the cognitive gain index of the user on the test questions is calculated.

[0044] The cognitive gain index is fed back to the knowledge state analysis model, and the parameters in the model are iterated.

[0045] By adopting the above technical solution, real-time capture of user answer response data (such as option selection sequence and draft trajectory) and high-precision problem-solving process data (including step delay distribution, modification traces, and auxiliary tool call records) is used to construct a holographic learning behavior map. Then, a dynamic time warping algorithm is used to compare the deviation between the actual answer path and the expected problem-solving path, accurately quantify the type of cognitive bias, and simultaneously calculate the cognitive gain index. This index integrates the improvement of cognitive load balance, expanding the dimension of training effect evaluation from a single indicator to a three-dimensional space. Finally, ΔG and path deviation data are fed back to the knowledge state analysis model to iterate the parameters in the model and ensure the real-time performance of the knowledge state analysis model.

[0046] Secondly, the above-mentioned inventive objective of this application is achieved through the following technical solutions:

[0047] An intelligent test question generation system based on learning behavior analysis, the intelligent test question generation system based on learning behavior analysis includes:

[0048] The multi-source learning behavior data acquisition module is used to acquire real-time interactive behavior data and performance data of learners in the user's online learning platform. The interactive behavior data includes the duration of time spent on knowledge points and the behavior of jumping to and from learning videos. The performance data includes answer records and the frequency of incorrect test questions.

[0049] The behavioral cognition association module is used to input the interactive behavior data and performance data into the pre-trained knowledge state analysis model to generate a user knowledge state matrix, which includes the knowledge point mastery degree, the type of wrong question and the cognitive load index.

[0050] The dynamic learning needs analysis module is used to identify the set of target knowledge points that need to be strengthened and the corresponding cognitive training type based on the knowledge state matrix and the preset teaching target library.

[0051] The test question generation module is used to dynamically assemble the question stem, distractors, and solution path hints based on the set of target knowledge points to be strengthened and the corresponding cognitive training type, and to generate personalized test questions by calling the test question element combination algorithm.

[0052] By adopting the above technical solution, multi-dimensional interactive behavior data (including knowledge point dwell time and video viewing / jumping behavior) and structured performance data (such as answer records and frequency of incorrect questions) from users' online learning platforms are captured in real time, constructing a fine-grained learning behavior profile. This solves the problem of traditional assessment relying on a single answer result. Furthermore, the fused data is input into a pre-trained knowledge state analysis model, which uses temporal behavior pattern recognition and error clustering algorithms to generate a user knowledge state matrix containing quantifiable values ​​of knowledge point mastery (accurate to the 0-1 range), error type classification labels (such as conceptual confusion / calculation error / logical gap), and a cognitive load index (calculated based on attention fluctuation characteristics). This significantly improves the accuracy of cognitive state assessment and overcomes the limitations of traditional assessment methods. Existing technologies neglect the implicit learning process. Based on the matrix, the achievement threshold of the teaching objective library is dynamically matched. The set of target knowledge points that need to be strengthened is located through knowledge gap topology analysis. The cognitive training type is adaptively selected according to the type of wrong questions (such as adversarial training for conceptual confusion and transfer training for logical gaps), realizing an intelligent decision-making closed loop from "knowledge gap detection" to "training strategy generation". Finally, the test question element combination algorithm is called to dynamically assemble the semantic framework of the question stem according to the core examination elements of the target knowledge point set. The deceptiveness of the interference items is optimized by using Generative Adversarial Network (GAN), and the problem-solving path prompts that are accurately matched with the user's historical error patterns are injected (such as inserting a knowledge point definition comparison table for conceptual confusion errors), generating completely personalized test questions.

[0053] Thirdly, the above-mentioned objectives of this application are achieved through the following technical solutions:

[0054] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the intelligent test question generation method based on learning behavior analysis described above.

[0055] Fourthly, the above-mentioned objectives of this application are achieved through the following technical solutions:

[0056] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described intelligent test question generation method based on learning behavior analysis.

[0057] In summary, this application includes at least one of the following beneficial technical effects:

[0058] 1. By capturing multi-dimensional interactive behavior data (including knowledge point dwell time and video viewing / jumping behavior) and structured performance data (such as answer records and frequency of incorrect questions) from users' online learning platforms in real time, a fine-grained learning behavior profile is constructed, solving the shortcomings of traditional assessments that rely on single answer results. Then, the fused data is input into a pre-trained knowledge state analysis model, which uses time-series behavior pattern recognition and error clustering algorithms to generate a user knowledge state matrix containing quantifiable values ​​of knowledge point mastery (accurate to the 0-1 range), error type classification labels (such as conceptual confusion / calculation error / logical gap), and a cognitive load index (calculated based on attention fluctuation characteristics). This significantly improves the accuracy of cognitive state assessment and overcomes the limitations of existing technologies. This addresses the issue of implicit learning processes being overlooked. Based on the matrix, the target threshold of the teaching objective library is dynamically matched. Through knowledge gap topology analysis, the set of target knowledge points that need to be strengthened is located. The cognitive training type is adaptively selected according to the type of incorrect questions (e.g., adversarial training for conceptual confusion and transfer training for logical gaps), realizing an intelligent decision-making closed loop from "knowledge gap detection" to "training strategy generation". Finally, the question element combination algorithm is called to dynamically assemble the semantic framework of the question stem according to the core examination elements of the target knowledge point set. The persuasiveness of the interference items is optimized by using Generative Adversarial Network (GAN), and the solution path prompts that are accurately matched with the user's historical error patterns are injected (e.g., inserting a knowledge point definition comparison table for conceptual confusion errors), generating completely personalized questions.

[0059] 2. By accurately capturing users' dynamic decision-making behavior during the question-answering process through event tracking logs (including the number of option modifications and the final submission delay), and simultaneously recording the confusion markers and related notes self-annotated by users in the error notebook, traditional performance data is extended to the cognitive decision-making level. This upgrades error attribution analysis from simple result judgment to process tracking (e.g., modification count > 3 times and delay > 2 times standard deviation are marked as "conceptual hesitation type" errors), significantly improving the accuracy of error pattern recognition. At the same time, pause / speed playback segments in video learning are extracted and bound to knowledge point tags (e.g., pause rate > 80% in key formula derivation segments is automatically associated with weak knowledge points), transforming unstructured interaction behavior into quantifiable attention allocation indicators. By fully utilizing video learning data, aggregating the above performance data and interaction behavior data, a user behavior vector sequence with millisecond-level timestamps is generated. Through time-series alignment technology, multi-dimensional heterogeneous data is integrated to construct a three-dimensional behavioral profile covering "question-answering decision chain - video focus - self-reflection text," providing a high-fidelity data foundation for subsequent knowledge state modeling.

[0060] 3. By decoupling multi-dimensional features from timestamped user behavior vector sequences, we separate knowledge proficiency features (calculated using exponential decay weighting based on answer accuracy), attention features (quantified by the coefficient of variation of video pause intervals), and metacognitive strategy features (using BERT clustering analysis of incorrect question annotation text). These decoupled features are then input into a pre-trained knowledge state analysis model. A graph neural network is used to dynamically construct a topological association strength matrix between knowledge points (e.g., the association strength between trigonometric functions and vector knowledge points), accurately capturing implicit knowledge dependency chains and addressing the deficiency of static knowledge graphs in reflecting individual cognitive differences. Based on this topological matrix, we integrate historical answer accuracy to generate dual-channel quantitative indicators: knowledge point mastery and cognitive load index. Simultaneously, an error pattern classifier is constructed based on historical answer data. Through confusion matrix analysis (e.g., similar option error rates >40% are marked as conceptual confusion), incorrect question types are refined into actionable training targets. Finally, these are integrated into a user knowledge state matrix, providing an irreplaceable cognitive map for subsequent target knowledge point positioning.

[0061] 4. In adversarial training, based on a confusing concept library (such as "stationary point" and "extreme point" in mathematics, and "electrolyte" and "non-electrolyte" in chemistry), distractors with semantic similarity >85% but belonging to different knowledge points are generated. Cosine similarity of word vectors is used to ensure the cognitive confusion between distractors and correct answers, and to address conceptual confusion errors in a targeted manner. In transfer training, contextualized distractors are generated by pre-set cross-disciplinary scenario templates (such as mapping mathematical derivatives to instantaneous velocity cases in physics, and transferring chemical equilibrium to ecological population competition models), which improves the recognition rate of abstract knowledge points. Then, adversarial generative networks are used to dynamically optimize the deceptiveness weight of distractors, calculate deceptiveness scores based on historical answer data, and filter out low-deceptive distractors. Attached Figure Description

[0062] Figure 1 This is a flowchart of an intelligent test question generation method based on learning behavior analysis in one embodiment of this application;

[0063] Figure 2 This is a flowchart illustrating the implementation of step S10 in an embodiment of the intelligent test question generation method based on learning behavior analysis in this application.

[0064] Figure 3 This is a flowchart illustrating the implementation of step S20 in an embodiment of the intelligent test question generation method based on learning behavior analysis in this application.

[0065] Figure 4 This is a flowchart illustrating the implementation of step S30 in an embodiment of the intelligent test question generation method based on learning behavior analysis in this application.

[0066] Figure 5 This is a flowchart illustrating the implementation of step S40 in an embodiment of the intelligent test question generation method based on learning behavior analysis in this application.

[0067] Figure 6 This is a flowchart illustrating the implementation of step S42 in an embodiment of the intelligent test question generation method based on learning behavior analysis in this application.

[0068] Figure 7 This is another implementation flowchart of the intelligent test question generation method based on learning behavior analysis in one embodiment of this application;

[0069] Figure 8 This is a principle block diagram of an intelligent test question generation system based on learning behavior analysis in one embodiment of this application;

[0070] Figure 9 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0071] The present application will be further described in detail below with reference to the accompanying drawings.

[0072] In one embodiment, such as Figure 1 As shown, this application discloses an intelligent test question generation method based on learning behavior analysis, which specifically includes the following steps:

[0073] S10: Real-time acquisition of learner interaction behavior data and performance data in the online learning platform. The interaction behavior data includes the duration of time spent on knowledge points and the behavior of jumping to and from learning videos. The performance data includes answer records and the frequency of incorrect test questions.

[0074] Specifically, by capturing multi-dimensional interactive behavior data (including the duration of time spent on knowledge points and video viewing and jumping behavior) and structured performance data (such as answer records and the frequency of incorrect questions) in real time on the user's online learning platform, a fine-grained learning behavior profile is constructed, which solves the shortcomings of traditional assessment that rely on a single answer result.

[0075] S20: Input the interaction behavior data and performance data into the pre-trained knowledge state analysis model to generate a user knowledge state matrix, which includes the degree of mastery of knowledge points, the type of wrong questions, and the cognitive load index.

[0076] Specifically, the model integrates data input into a pre-trained knowledge state analysis model and uses temporal behavior pattern recognition and error clustering algorithms to generate a user knowledge state matrix that includes quantified values ​​of knowledge point mastery (accurate to the 0-1 range), error type classification labels (such as conceptual confusion, calculation error, and logical gap), and cognitive load index (calculated based on attention fluctuation characteristics). This improves the accuracy of cognitive state assessment and significantly overcomes the problem of existing technologies ignoring the implicit learning process.

[0077] S30: Based on the knowledge state matrix and combined with the preset teaching target library, identify the set of target knowledge points that need to be strengthened and the corresponding cognitive training type.

[0078] Specifically, based on the matrix, the target threshold of the teaching objective library is dynamically matched, the set of target knowledge points that need to be strengthened is located through knowledge vulnerability topology analysis, and the cognitive training type is adaptively selected according to the type of wrong questions (such as adversarial training for conceptual confusion and transfer training for logical gaps), so as to realize the intelligent decision-making closed loop from "knowledge vulnerability detection" to "training strategy generation".

[0079] S40: Based on the set of target knowledge points to be strengthened and the corresponding cognitive training type, call the question element combination algorithm to dynamically assemble the question stem, distractors and solution path hints to generate personalized questions.

[0080] Specifically, the algorithm for combining test elements is invoked to dynamically assemble the semantic framework of the question stem based on the core examination elements of the target knowledge point set. Generative Adversarial Network (GAN) is used to optimize the persuasiveness of the distractors, and problem-solving path hints that are precisely matched with the user's historical error patterns are injected (such as inserting a knowledge point definition comparison table for conceptual confusion errors) to generate fully personalized test questions.

[0081] In this embodiment, by capturing multi-dimensional interactive behavior data (including knowledge point dwell time and video viewing / jumping behavior) and structured performance data (such as answer records and frequency of incorrect questions) from users' online learning platforms in real time, a fine-grained learning behavior profile is constructed, solving the deficiency of traditional assessment relying on a single answer result. Then, the fused data is input into a pre-trained knowledge state analysis model, and time-series behavior pattern recognition and error clustering algorithms are used to generate a user knowledge state matrix containing quantified values ​​of knowledge point mastery (accurate to the 0-1 range), error type classification labels (such as conceptual confusion / calculation error / logical gap), and a cognitive load index (calculated based on attention fluctuation characteristics). This improves the accuracy of cognitive state assessment and significantly overcomes the problem of existing technologies ignoring implicit learning processes. Based on this matrix, dynamic... The system matches the threshold of the teaching objective database, locates the set of target knowledge points that need to be strengthened through knowledge gap topology analysis, and adaptively selects cognitive training types based on the type of wrong questions (such as adversarial training for conceptual confusion and transfer training for logical gaps), realizing an intelligent decision-making closed loop from "knowledge gap detection" to "training strategy generation". Finally, it calls the question element combination algorithm to dynamically assemble the semantic framework of the question stem according to the core examination elements of the target knowledge point set, uses adversarial generative network (GAN) to optimize the persuasiveness of the distractors, and injects problem-solving path prompts that accurately match the user's historical error patterns (such as inserting a knowledge point definition comparison table for conceptual confusion errors), generating fully personalized questions, realizing a closed-loop intelligent teaching from user behavior analysis to targeted training, and dynamically generating personalized questions adapted to individual cognitive gaps.

[0082] In one embodiment, such as Figure 2 As shown, in step S10, which involves acquiring user interaction behavior data and performance data from the online learning platform in real time, the specific steps include:

[0083] S11: Based on the tracking logs, capture the number of times users modify their answer choices and the final submission delay, record the confusion marks and related notes that users independently annotate in the error notebook, and integrate them to form performance data.

[0084] Specifically, by accurately capturing users' dynamic decision-making behavior during the question-answering process through data logging (including the number of option modifications and the final submission delay), and simultaneously recording the confusion marks and related notes self-marked by users in the error notebook, traditional performance data is extended to the cognitive decision-making level, upgrading error attribution analysis from simple result judgment to process tracking (e.g., the number of modifications > 3 and the delay > 2 times the standard deviation are marked as "conceptual hesitation type" vulnerabilities), significantly improving the accuracy of error pattern recognition.

[0085] S12: Extract pause / speed playback segments from video learning and their associated knowledge point tags to generate interactive behavior data.

[0086] Specifically, pause / speed playback segments in video learning are extracted and bound to knowledge point tags (e.g., pause rate > 80% in key formula derivation segments is automatically associated with weak knowledge points), transforming unstructured interactive behaviors into quantifiable attention allocation indicators, and making full use of video learning data.

[0087] S13: Aggregate the performance data and interaction behavior data to form a time-stamped user behavior vector sequence.

[0088] Specifically, by aggregating the aforementioned performance data and interaction behavior data, a user behavior vector sequence with millisecond-level timestamps is generated. Multidimensional heterogeneous data is then integrated using time-series alignment technology to construct a three-dimensional behavioral profile covering "answer decision chain - video focus - self-reflection text," providing a high-fidelity data foundation for subsequent knowledge state modeling.

[0089] In one embodiment, such as Figure 3 As shown, in step S20, the interaction behavior data and performance data are input into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, specifically including:

[0090] S21: Decouple the user behavior vector sequence with timestamps by separating the knowledge proficiency features, attention features, and metacognitive strategy features.

[0091] Specifically, the user behavior vector sequence with timestamps is decoupled from multiple dimensions to separate knowledge proficiency features (calculated by exponential decay weighting based on the correct answer rate), attention features (quantified by the coefficient of variation of video pause intervals), and metacognitive strategy features (using BERT clustering analysis of incorrect question annotation text).

[0092] S22: Input the knowledge proficiency features, attention features, and metacognitive strategy features into the pre-trained knowledge state analysis model to construct a topological correlation strength matrix between knowledge points.

[0093] Specifically, the decoupled features are input into the pre-trained knowledge state analysis model, and graph neural networks are used to dynamically construct the topological correlation strength matrix between knowledge points (such as the correlation strength between trigonometric functions and vector knowledge points) to accurately capture implicit knowledge dependency chains and solve the defect that static knowledge graphs cannot reflect individual cognitive differences.

[0094] S23: Based on the topological correlation strength matrix between knowledge points and the user's historical answer accuracy, calculate the quantitative indicators of each dimension in the topological correlation strength matrix between knowledge points, and generate the knowledge point mastery and cognitive load index.

[0095] Specifically, based on this topological matrix, the historical answer accuracy is integrated to generate dual-channel quantitative indicators: knowledge point mastery and cognitive load index, thereby reducing the error rate of cognitive state modeling.

[0096] S24: Obtain error pattern classification based on the user's historical answer accuracy rate, determine the type of wrong question based on the error pattern classification, and generate the user's knowledge state matrix based on the knowledge point mastery, cognitive load index and wrong question type.

[0097] Specifically, an error pattern classifier is constructed based on historical answer data. Through confusion matrix analysis (e.g., similar options with an error rate >40% are marked as conceptual confusion type), the types of incorrect questions are refined into actionable training targets. Finally, these are integrated into a user knowledge state matrix, providing an irreplaceable cognitive map for subsequent target knowledge point positioning.

[0098] In one embodiment, such as Figure 4 As shown, in step S30, based on the knowledge state matrix and combined with the preset teaching objective library, the set of target knowledge points that need to be strengthened and the corresponding cognitive training type are identified, specifically including:

[0099] S31: Compare the knowledge state matrix with the achievement thresholds in the teaching objective base to locate the knowledge points that have not met the standards.

[0100] Specifically, the system intelligently compares the mastery level of knowledge points in the knowledge status matrix with the preset threshold of the teaching objective database (e.g., the mastery level of the junior high school mathematics function chapter is required to be ≥0.8), accurately locates the knowledge points that have not met the standard, and avoids the omissions caused by traditional manual screening.

[0101] S32: Select the knowledge point training type based on the error type in the knowledge state matrix, and determine the training intensity level based on the cognitive load index in the knowledge state matrix.

[0102] S33: Based on the knowledge point training type and training intensity level, output a set of required instructions containing the target knowledge point ID, training type, and intensity level.

[0103] Specifically, based on the detailed error types in the matrix, the cognitive training type is dynamically selected: conceptual confusion errors trigger adversarial training (generating similar concept distractors), calculation error errors activate step-by-step verification training (embedding automated verification anchors), and logical discontinuity errors initiate reasoning chain reconstruction training (injecting a visual framework). This achieves precise matching between the root cause of the error and the correction strategy. At the same time, the training intensity level is determined by combining the cognitive load index, and the complexity of the test questions is dynamically adjusted (such as the number of implicit conditions in the question stem increasing or decreasing linearly with the intensity coefficient), effectively avoiding the risk of cognitive overload, and finally outputting a structured demand instruction set.

[0104] In one embodiment, such as Figure 5As shown, in step S40, based on the set of target knowledge points to be strengthened and the corresponding cognitive training type, the test element combination algorithm is invoked to dynamically assemble the question stem, distractors, and solution path hints to generate personalized test questions, specifically including:

[0105] S41: Match the core examination elements from the test question material library according to the target knowledge point ID of the required instruction set.

[0106] Specifically, based on the target knowledge point ID of the demand instruction set, the core examination elements are intelligently matched from the test question material library (such as binding the three elements of vertex form, axis of symmetry, and discriminant form for the knowledge point of "quadratic function extrema") to ensure the targeting of test questions.

[0107] S42: Determine the data characteristics based on the training set, determine the machine learning algorithm based on the data characteristics, and train the preset initial machine learning model based on the machine learning algorithm and the training set.

[0108] S43: Determine the question complexity parameter based on the training intensity level, and based on the question complexity parameter and question distractors, inject solution path hints that match the user's wrong question type to generate the final personalized question.

[0109] Specifically, the strategy of generating distractors is dynamically activated for different training types. For example, in adversarial training, a confusion concept library is called to generate distractors with a semantic similarity greater than 85% (such as replacing "extreme point" with "stationary point"). In transfer training, the distractors are mapped to cross-disciplinary scenario templates (such as using physical kinematics cases to package mathematical function problems) to increase their deceptiveness.

[0110] Furthermore, the complexity parameters of the test questions are precisely adjusted according to the training intensity level to effectively balance the training challenge. Finally, problem-solving path prompts that strictly match the user's wrong question types are injected to generate fully structured personalized test questions.

[0111] In one embodiment, such as Figure 6 As shown, in step S42, which involves selecting a distractor generation strategy based on the training type and outputting distractor items in response to the distractor generation strategy, the specific steps include:

[0112] S421: The training types include adversarial training and transfer training. In adversarial training, interference items with similar semantics but different knowledge point affiliations are generated based on the concept of confusion.

[0113] S422: In transfer training, core knowledge points are mapped to preset interdisciplinary scenario templates to generate interference items.

[0114] Specifically, in adversarial training, distractors with semantic similarity >85% but belonging to different knowledge points are generated based on a confusing concept library (such as "stationary point" and "extreme point" in mathematics, and "electrolyte" and "non-electrolyte" in chemistry). Cosine similarity of word vectors is used to ensure the cognitive confusion between the distractors and the correct answer, and to address conceptual confusion errors in a targeted manner. In transfer training, contextualized distractors are generated by pre-set cross-disciplinary scenario templates (such as mapping mathematical derivatives to instantaneous velocity cases in physics, and transferring chemical equilibrium to ecological population competition models), which improves the recognition rate of abstract knowledge points.

[0115] S423: Optimize the deceptiveness weights of distractors and correct answers using an adversarial generative network, and filter out low-deceptive distractors based on their weights.

[0116] Specifically, an adversarial generative network is used to dynamically optimize the deceptiveness weights of interference items, calculate the deceptiveness score based on historical answer data, and filter out low-deceptive interference items.

[0117] In one embodiment, such as Figure 7 As shown, intelligent test question generation methods based on learning behavior analysis also include:

[0118] S50: Obtain the user's answer response dataset to the generated personalized test questions, which contains the problem-solving process data.

[0119] S60: Compare the deviation between the actual answer path and the expected answer path to calculate the user's cognitive gain index of the test questions.

[0120] S70: Feedback the cognitive gain index to the knowledge state analysis model and iterate the parameters in the model.

[0121] Specifically, real-time capture of user answer response data (such as option selection sequence and draft trajectory) and high-precision problem-solving process data (including step delay distribution, modification traces, and auxiliary tool call records) is used to construct a holographic learning behavior map. Then, a dynamic time warping algorithm is used to compare the deviation between the actual answer path and the expected problem-solving path, accurately quantify the type of cognitive bias, and simultaneously calculate the cognitive gain index: ΔG = (score of this answer - historical average score) × cognitive load index;

[0122] This index enhances the fusion capability and cognitive load balance, expanding the evaluation dimension of training effect from a single indicator to a three-dimensional space. Ultimately, it feeds back ΔG and path deviation data to the knowledge state analysis model, iterating the parameters in the model and ensuring the real-time performance of the knowledge state analysis model.

[0123] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0124] In one embodiment, an intelligent test question generation system based on learning behavior analysis is provided, which corresponds one-to-one with the intelligent test question generation method based on learning behavior analysis in the above embodiments. For example... Figure 8 As shown, this intelligent test question generation system based on learning behavior analysis includes a multi-source learning behavior data acquisition module, a behavior-cognition association module, a dynamic learning needs analysis module, and a test question generation module. Detailed descriptions of each functional module are as follows:

[0125] The multi-source learning behavior data acquisition module is used to acquire real-time interactive behavior data and performance data of learners in the user's online learning platform. The interactive behavior data includes the duration of time spent on knowledge points and the behavior of jumping to and from learning videos. The performance data includes answer records and the frequency of incorrect test questions.

[0126] The behavioral cognition association module is used to input the interactive behavior data and performance data into the pre-trained knowledge state analysis model to generate a user knowledge state matrix, which includes the knowledge point mastery degree, the type of wrong question and the cognitive load index.

[0127] The dynamic learning needs analysis module is used to identify the set of target knowledge points that need to be strengthened and the corresponding cognitive training type based on the knowledge state matrix and the preset teaching target library.

[0128] The test question generation module is used to dynamically assemble the question stem, distractors, and solution path hints based on the set of target knowledge points to be strengthened and the corresponding cognitive training type, and to generate personalized test questions by calling the test question element combination algorithm.

[0129] Specific limitations regarding the intelligent test question generation system based on learning behavior analysis can be found in the limitations of the intelligent test question generation method based on learning behavior analysis mentioned above, and will not be repeated here. Each module in the aforementioned intelligent test question generation system based on learning behavior analysis can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in an electronic device, or stored in the memory of an electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0130] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, this electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores monitoring data from multiple patients and diabetes analysis models. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements an intelligent test item generation method based on learning behavior analysis.

[0131] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0132] Real-time acquisition of learner interaction behavior data and performance data in the online learning platform, wherein the interaction behavior data includes the duration of time spent on knowledge points and the behavior of jumping to and from learning videos, and the performance data includes answer records and the frequency of incorrect questions;

[0133] The interactive behavior data and performance data are input into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, which includes the degree of mastery of knowledge points, the type of wrong questions, and the cognitive load index.

[0134] Based on the knowledge state matrix and combined with the preset teaching objective library, identify the set of target knowledge points that need to be strengthened and the corresponding cognitive training type.

[0135] Based on the set of target knowledge points to be strengthened and the corresponding cognitive training types, the test element combination algorithm is invoked to dynamically assemble the question stem, distractors, and solution path hints to generate personalized test questions.

[0136] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0137] Real-time acquisition of learner interaction behavior data and performance data in the online learning platform, wherein the interaction behavior data includes the duration of time spent on knowledge points and the behavior of jumping to and from learning videos, and the performance data includes answer records and the frequency of incorrect questions;

[0138] The interactive behavior data and performance data are input into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, which includes the degree of mastery of knowledge points, the type of wrong questions, and the cognitive load index.

[0139] Based on the knowledge state matrix and combined with the preset teaching objective library, identify the set of target knowledge points that need to be strengthened and the corresponding cognitive training type.

[0140] Based on the set of target knowledge points to be strengthened and the corresponding cognitive training types, the test element combination algorithm is invoked to dynamically assemble the question stem, distractors, and solution path hints to generate personalized test questions.

[0141] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0142] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0143] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An intelligent test question generation method based on learning behavior analysis, characterized in that, The intelligent test question generation method based on learning behavior analysis comprises the following steps: Real-time acquisition of interactive behavior data and performance data of learners in a user online learning platform, wherein the interactive behavior data comprises knowledge point stay duration and learning video viewing jump behavior, and the performance data comprises answer recording and error test question frequency; Input of the interactive behavior data and performance data into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, wherein the user knowledge state matrix comprises knowledge point mastery, error test question type and cognitive load index; Based on the knowledge state matrix, in combination with a preset teaching target library, identification of a current target knowledge point set to be strengthened and a corresponding cognitive training type, specifically comprising: Comparison of the knowledge state matrix with a standard threshold in the teaching target library to locate a non-standard knowledge point; Selection of a knowledge point training type according to the error test question type in the knowledge state matrix, and determination of a training intensity level based on the cognitive load index in the knowledge state matrix; Based on the knowledge point training type and the training intensity level, output of a requirement instruction set comprising a target knowledge point ID, a training type and an intensity level; Based on the target knowledge point set to be strengthened and the corresponding cognitive training type, calling of a test question element combination algorithm to dynamically assemble a stem, a distractor and a problem solving path prompt to generate an individualized test question, specifically comprising: Matching of a core examination element from a test question material library according to the target knowledge point ID of the requirement instruction set; Based on the training type, selection of a distractor generation strategy, and output of a test question distractor in response to the distractor generation strategy, specifically comprising: The training type comprises adversarial training and transfer training. In adversarial training, a distractor with similar semantics but different knowledge point attribution is generated based on a confused concept. In transfer training, a distractor is generated by mapping a core knowledge point to a preset interdisciplinary scenario template. The degree of confusion weight of the distractor and the correct answer is optimized through an adversarial generation network, and the distractor with low confusion is selected according to the weight of the distractor and the correct answer. According to the training intensity level, a test question complexity parameter is determined, and a problem solving path prompt matching the user's error test question type is injected based on the test question complexity parameter and the test question distractor to generate a final individualized test question. 2.The intelligent test question generation method based on learning behavior analysis according to claim 1, wherein, The real-time acquisition of interactive behavior data and performance data in the user online learning platform specifically comprises: Based on the buried point log, the number of option modifications during the user's answering and the final submission delay are captured, the confusion marks and associated note texts marked by the user in the error test question book are recorded, and the performance data is integrated to form the performance data; The pause / speed play segments in video learning and their associated knowledge point labels are extracted to generate interactive behavior data; The performance data and interactive behavior data are aggregated to form a time-stamped user behavior vector sequence. 3.The intelligent test question generation method based on learning behavior analysis according to claim 2, characterized in that, The interactive behavior data and performance data are input into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, specifically comprising: Feature decoupling is performed on the time-stamped user behavior vector sequence to separate knowledge proficiency features, attention features and metacognitive strategy features; Input the knowledge proficiency feature, attention feature and metacognitive strategy feature into a pre-trained knowledge state analysis model to build a topological correlation strength matrix between knowledge points; Based on the topological correlation strength matrix between knowledge points and the historical answering accuracy of the user, calculate the quantization indicators of each dimension in the topological correlation strength matrix between knowledge points to generate a knowledge point mastery degree and a cognitive load index; According to the historical answering accuracy of the user, obtain an error pattern classification, determine a wrong question type based on the error pattern classification, and generate a user knowledge state matrix based on the knowledge point mastery degree, cognitive load index and wrong question type. 4.The intelligent test question generation method based on learning behavior analysis of claim 1, wherein, After the personalized test question is generated by dynamically assembling the stem, interference item and problem solving path prompt based on the target knowledge point set to be strengthened and the corresponding cognitive training type, the intelligent test question generation method based on learning behavior analysis further comprises: Obtain the problem solving process data of the user's answering response data set of the generated personalized test question; Compare the deviation degree of the actual answering path and the expected problem solving path to calculate the cognitive gain index of the user for the test question; According to the cognitive gain index, feedback to the knowledge state analysis model to iterate the parameters in the model.

5. An intelligent test question generation system based on learning behavior analysis, characterized by, The intelligent test question generation system based on learning behavior analysis comprises: A multi-source learning behavior data acquisition module is used to acquire the interactive behavior data and achievement data of learners in the online learning platform of the user in real time, the interactive behavior data includes knowledge point stay duration and learning video viewing and jumping behavior, and the achievement data includes answering record and error test question frequency; A behavior-cognition correlation module is used to input the interactive behavior data and achievement data into a pre-trained knowledge state analysis model to generate a user knowledge state matrix, the user knowledge state matrix includes a knowledge point mastery degree, a wrong question type and a cognitive load index; A dynamic learning demand analysis module is used to identify a target knowledge point set to be strengthened and a corresponding cognitive training type based on the knowledge state matrix and a preset teaching target library, specifically including: comparing the knowledge state matrix with the passing threshold in the teaching target library to locate the non-passing knowledge points; selecting a knowledge point training type according to the wrong question type in the knowledge state matrix, determining a training intensity level based on the cognitive load index in the knowledge state matrix; based on the knowledge point training type and the training intensity level, output a demand instruction set including the target knowledge point ID, the training type and the intensity level; The test question generation module is configured to call a test question element combination algorithm according to the target knowledge points set to be strengthened and the corresponding cognitive training type, dynamically assemble a stem, a distractor and a problem solving path prompt, and generate a personalized test question, and specifically includes: matching core examination elements from a test question material library according to the target knowledge point ID of the requirement instruction set; selecting a distractor generation strategy based on the training type, outputting a test question distractor in response to the distractor generation strategy, and specifically including: the training type includes adversarial training and transfer training, in the adversarial training, generating a distractor with similar semantics but different knowledge point attributions based on a confused concept; in the transfer training, mapping the core knowledge point to a preset cross-disciplinary scene template to generate a distractor; optimizing the confusion degree weight of the distractor and the correct answer through an adversarial generation network, screening a low-confusion distractor according to the weight of the distractor and the correct answer; determining a test question complexity parameter according to a training intensity level, injecting a problem solving path prompt matched with the user's wrong question type based on the test question complexity parameter and the test question distractor, and generating a final personalized test question.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the intelligent test question generation method based on learning behavior analysis according to any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, wherein the computer program comprises the following steps of: receiving a request for a resource from a client; determining whether the client is authorized to access the resource; and if the client is authorized to access the resource, providing the resource to the client. The computer program is executed by the processor to implement the steps of the intelligent test question generation method based on learning behavior analysis according to any one of claims 1 to 4.

Images